Forecasting the Term Structure in Emerging Markets using Extensions of the Dynamic Nelson-Siegel Model

Abstract

The dynamic Nelson-Siegel model and its extensions are used by many central banks to forecast the term structure. Their forecasting performance has been studied for many countries, but a little can be said about their accuracy for emerging market economies (EMEs). In this work we test the traditional dynamic Nelson-Siegel models, their extensions without the “curvature” factor and with inflation for six EMEs. Other extensions are model selection via BIC minimization and the Bayesian estimation with the conjugate Normal-inverse Wishart prior, which are novel in the field. The results indicate that inflation data and the Bayesian approach improve the forecasting performance relative to the traditional models. We also conclude that the multivariate dynamic Nelson-Siegel models often outperform univariate ones, while the “curvature” factor is rarely helpful for forecasting at a long horizon.